In an increasingly IoT implementation with AI functionality, good balance of power consumption and compute performance are often a requirement for the embedded system. Aetina brings more performance per watt for Embedded and IoT, using Nvidia AGX Xavier module with Aetina's Nvidia Jetson carrier board - AX710 to form a high-performance edge computing platform, and imported the open source library - Openpose algorithm, successfully setting up the application of pose estimation. This live demo used the skeleton detection through user's motion and instructed the robotic to generate feedback. Furthermore, AGX Xavier provides 4 times better compute efficiency and 1.5 times on AI inference, which perfectly fit the edge devices in AI pose estimation projects.
▌AGX Xavier / RTX 2080 Comparison
AGX Xavier provides 4 times better compute efficiency and 1.5 times on AI inference than RTX 2080
The hardware of the pose estimation application uses the Nvidia AGX Xavier computing module and Aetina’s Jetson carrier board-AX710, which provides 512 CUDA cores, 64 Tensor cores and 11 TFLOPS AI inference performance to the AI project. Compared with Nvidia RTX 2080 high-performance graphics card, RTX 2080 provides 20.14 TFLOPS of compute performance, but the power requirement is 225W. In the other hand, AGX Xavier’s power consumption is only 30W, bringing 0.367 (TFLOPS/W) inference-power-ratio, which provides nearly three times more efficient than RTX 2080’s 0.09 (TFLOPS/W). In addition, Aetina’s AGX Xavier computing platform supports a wide range of I/O extension modules, with the ability to expand high-speed LAN cards and a variety of digital or analog cameras.
▌Pose Estimation by Openpose & Aetina Xavier Platform
Real-time multiple people detection for pose estimation
In terms of software, we used the Openpose algorithm of the open source library to detect the skeleton of the human body through grabbing feature points. The feature of this algorithm is that it is able to detect multiple people in real time. As long as the skeleton data is collected and the poses are defined, additional scripts can be embedded after the motion is detected for more applications. In this application, we set the platform to automatically deliver a business card as soon as it detects a specific hand movement. In the same way, with different actions and instructions, developers can use this high inference-consumption-ratio platform to fulfill many AI applications, such as fall detection in smart medical and customer behavior analysis of smart retail.
|Contact Us||Product Inquiry||Relevant Information|
|Aetina Jetson AGX Xavier Carrier Board|
|High-speed Lan Supported Carrier Board for Nvidia Jetson AGX Xavier|
|112mm x 107mm|
|-25°C to +80°C|
|Support extension M.2/high-speed LAN card/mPCIe device|